“And that’s acknowledging the fact that LLMs — even those developed a legion of well-resourced and well-fed PHDs, like GPT-4, and based on a well-curated dataset of high-quality material — have a habit of hallucinating facts, figures, and events.”
Our reading, the sentence that was checked:
LLMs, even well-resourced ones like GPT-4 trained on curated data, habitually hallucinate facts, figures, and events.
Assessment: Supported Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Elon's Gordian Knot, paragraph 21.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-elons-gordian-knot-31dd8b-a1
We recorded 12 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2023-12-05. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- even well-resourced models trained on curated data (qualifier, stated by the speaker)
From the speaker
- LLMs
- even those
- like GPT-4
- have a habit of hallucinating facts, figures, and events
Our interpretation
- 'Have a habit' means a recurring, demonstrated failure mode, not that most answers are fabricated.
- GPT-4 is the named example of a well-resourced model; the sentence does not assert a measurable universal standard for training-data quality.
- For this ongoing technical observation, assess the GPT-4-era reporting period from its public introduction through the post date.
Counts as supported if: Evaluations demonstrate repeated factual fabrication by GPT-4 and other contemporary LLMs, despite training and measures intended to improve factuality.
Counts as contradicted if: Reliable evaluations establish that the alleged recurring fabrication did not occur in the relevant models, or that the cited examples were misattributed or not factual errors.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Both the developer's disclosure and independent research establish factual hallucination as a recurring problem in contemporary LLMs, including GPT-4. This supports the ordinary meaning of 'have a habit.' It does not establish that most answers were false or that every model had the same error rate.
Limitations
- The research uses constructed evaluation datasets, not a representative sample of everyday use.
- The evidence does not independently audit the quality of GPT-4's entire training dataset.
Evidence (2 items)
Supports the claim: OpenAI,
not fully reliable
Why it matters here: OpenAI acknowledges factual hallucinations in GPT-4. Its accompanying report of improvement over earlier models does not remove that recurring limitation.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: arXiv; Muru Zhang and colleagues,
their own mistakes
Why it matters here: The study reports repeated incorrect answers and false explanatory claims across three constructed datasets. The quoted result measures recognition of those errors in separate questioning, not the models' overall hallucination rate.
- Publisher
- arXiv; Muru Zhang and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Elon's Gordian Knot
And that’s acknowledging the fact that LLMs — even those developed a legion of well-resourced and well-fed PHDs, like GPT-4, and based on a well-curated dataset of high-quality material — have a habit of hallucinating facts, figures, and events.
- Where
- paragraph 21
- Published
- 2023-12-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-elons-gordian-knot-31dd8b-a1: Current. Outcome: Supported. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.